Big data-based teaching quality evaluation method and system

Abstract

The present application relates to the technical field of teaching evaluation, in particular to a teaching quality evaluation method and system based on big data, comprising the following steps: obtaining a teaching task performance point sequence and a target performance interval, constructing a multi-dimensional performance reference vector, calculating the difference and generating a target offset measurement parameter, constructing an embedded graph and generating a behavior transition sparsity distribution, extracting a behavior block to generate a density projection curve and intercepting a breaking point, constructing an evaluation vector and determining the grade. The present application extracts continuous teaching cycle data and constructs a multi-dimensional reference vector in combination with a teaching unit target performance interval, extracts a teaching stability breaking point, maps multiple key features to a high-dimensional space to construct an evaluation vector and compares the deviation with a standard state, realizes multi-dimensional dynamic modeling and stability analysis of teaching quality, and effectively improves the systematicness, dynamics and objectivity of teaching quality evaluation in complex teaching scenarios.

Description

Technical Field

[0001] This invention relates to the field of teaching evaluation technology, and in particular to a teaching quality evaluation method and system based on big data. Background Technology

[0002] The field of teaching evaluation technology involves assessing core aspects such as the effectiveness of teaching activities, the quality of the teaching process, teachers' teaching behaviors, and students' learning performance. It aims to achieve objective analysis and scientific judgment of teaching quality through quantitative and qualitative methods. This field typically encompasses the construction of evaluation indicator systems, teaching data collection and statistical analysis, and teaching outcome feedback mechanisms. Traditional teaching evaluation methods are mostly based on manual scoring, questionnaires, and classroom observation, which suffer from problems such as limited data sources, narrow evaluation dimensions, and strong subjectivity, making it difficult to comprehensively reflect the actual teaching effectiveness. With the development of information technology in education, big data technology has been gradually introduced into the teaching evaluation process. By centrally processing multi-source heterogeneous teaching data, it provides data support for evaluation models, enabling the quantification, refinement, and intelligence of evaluation methods.

[0003] Traditional teaching quality assessment methods rely on indicators such as student exam scores, teacher instruction, class attendance records, and classroom participation, using statistical summarization and average comparison to judge teaching quality. These methods typically involve manually collecting paper or electronic transcripts, compiling timetables, observing student classroom performance, distributing and collecting paper or online questionnaires, and weighting various indicators according to pre-set formulas to form an assessment conclusion. These methods depend on limited sample data, and the assessment process is mostly linear and static, lacking the ability to analyze large-scale, multi-dimensional, and dynamic data, making them unsuitable for systematically assessing teaching quality in complex teaching scenarios. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a teaching quality assessment method and system based on big data.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a teaching quality evaluation method based on big data, comprising the following steps:

[0006] S1: Obtain the original performance point value sequence of teaching tasks and the target performance interval of teaching units; extract continuous teaching cycle data from the original performance point value sequence of teaching tasks and construct a multi-dimensional performance reference vector for the same period by combining it with the target performance interval of teaching units.

[0007] S2: Difference the original performance point value sequence of the teaching task with the multidimensional performance reference vector of the same period, calculate the discrete variation amplitude within the period using the dynamic time warping algorithm, and generate the target offset measurement parameter by calculating the ratio of the discrete variation amplitude within the period to the target performance interval of the teaching unit.

[0008] S3: Construct a teaching activity embedding graph structure based on the original performance point value sequence of teaching tasks, calculate the co-occurrence probability difference between adjacent nodes, and accumulate the co-occurrence probability difference exceeding the behavioral coherence breaking threshold to generate behavioral transfer sparsity distribution information.

[0009] S4: Extract continuous behavior block sequences from the embedded graph structure of the teaching activities and calculate the Euclidean distance to generate behavior density projection curves. Use density-based noise to apply spatial clustering algorithms and extract stability breakpoints based on the behavior density projection curves.

[0010] S5: Construct a teaching quality evaluation vector based on the target offset metric parameters, behavior transition sparsity distribution information, and stability breakpoint, and calculate the deviation between the teaching quality evaluation vector and the standard stable state vector to determine the teaching quality level.

[0011] The present invention improves upon the following: the multi-dimensional performance reference vector of the same period includes a lower bound component of the performance benchmark, the mean of the expected target achievement, and a periodic fluctuation tolerance threshold; the target offset measurement parameters include positive and negative deviation polarity indicators, relative fluctuation amplitude ratio, and cumulative drift trend index; the behavior transfer sparsity distribution information includes the inter-node transfer probability density, path breakage cumulative weight, and sparse interval span distribution; the stability breakpoint includes the mutation occurrence timestamp index, density gradient jump amplitude, and behavior pattern conversion type label; and the teaching quality level includes a comprehensive stability score, teaching target achievement category, and abnormal risk warning level.

[0012] The present invention is improved in that the specific steps for obtaining the multi-dimensional performance reference vector of the same period are as follows:

[0013] S111: Retrieve the original performance point value sequence of teaching tasks, including historical scoring records, and the preset target performance interval of teaching units from the teaching management database. Analyze the timestamp attribute carried by each discrete data item in the original performance point value sequence of teaching tasks one by one. Sort the original performance point value sequence of teaching tasks in ascending order of time dimension according to the value of the timestamp attribute. Establish the corresponding data index mapping relationship between the original performance point value sequence of teaching tasks and the target performance interval of teaching units, and generate a combination of basic data for teaching tasks with timestamp attributes.

[0014] S112: Based on the combination of teaching task basic data with timestamp attributes, calculate the time interval value between two adjacent timestamps in the sequence, compare the time interval value with the preset teaching cycle continuity judgment threshold, if the time interval value is less than or equal to the teaching cycle continuity judgment threshold, then the corresponding data node is judged as a continuous node, identify and extract the time sequence segment composed of continuous nodes, remove isolated time point data that do not meet the continuity requirements, and obtain a continuous teaching cycle performance data subsequence.

[0015] S113: Call the target performance interval of the teaching unit in the combination of the continuous teaching cycle performance data subsequence and the teaching task basic data with timestamp attribute, set the lower and upper bounds of the target performance interval of the teaching unit as the reference axis of the multi-dimensional reference coordinate system, project each performance point value in the continuous teaching cycle performance data subsequence to the multi-dimensional reference coordinate system, calculate the Euclidean space position coordinate and vertical distance parameter of each performance point value relative to the reference axis, and vectorize and concatenate the position coordinate and vertical distance parameter according to the chronological order of the time dimension to construct the multi-dimensional performance reference vector of the same period.

[0016] The present invention is improved in that the step of obtaining the target offset metric parameter is specifically as follows:

[0017] S211: Perform a point-by-point difference operation based on time index on the original performance point value sequence of the teaching task and the multidimensional performance reference vector of the same period, calculate the numerical deviation between the two at the same time node, and generate an instantaneous performance deviation sequence.

[0018] S212: Based on the instantaneous performance deviation sequence, construct the minimum cumulative distance path between the original performance point value sequence of the teaching task and the multidimensional performance reference vector of the same period, calculate the overall fluctuation intensity of the sequence by integrating the path regularization cost and local deviation characteristics, and calculate the discrete variation amplitude within the period.

[0019] S213: Analyze the upper and lower boundary values ​​of the target performance interval of the teaching unit and calculate the interval span amplitude. Calculate the ratio of the discrete variation amplitude within the period to the interval span amplitude to obtain the target offset measurement parameter.

[0020] The present invention is improved in that the formula for obtaining the discrete variation amplitude within the period is specifically as follows:

[0021] ;

[0022] in, This represents the magnitude of discrete changes within the period. The sequence length representing the instantaneous performance deviation sequence. This represents the normalized instantaneous performance deviation value at the k-th time point in the instantaneous performance deviation sequence. This represents the weighting factor that decreases over time at the k-th time point. The normalized dynamic programming approach minimizes the cumulative path cost between the original performance point sequence of a teaching task and the multidimensional performance reference vector within the same period. The normalized performance value of the teaching task at the k-th time point in the original performance point sequence of the teaching task represents the teaching task performance value. This represents the normalized intra-period reference value at the k-th time point in the intra-period multi-dimensional performance reference vector. This represents the normalized adjustment coefficient for path cost.

[0023] The present invention is improved in that the step of obtaining the behavior transition sparsity distribution information is specifically as follows:

[0024] S311: Based on the original performance point value sequence of teaching tasks, parse the teaching activity type identifier corresponding to each performance point value, instantiate the teaching activity type identifier into an independent node in the topology network, establish directed connection edges according to the adjacent order of the teaching activity type identifier in the time dimension, and construct the teaching activity embedding graph structure using the reciprocal of the time interval between nodes as the edge weight parameter.

[0025] S312: For the embedded graph structure of the teaching activity, traverse all combinations of predecessor and successor nodes connected by directed edges, count the joint occurrence frequency of each node combination in the overall teaching cycle, convert it into conditional transition probability value, calculate the absolute value of the difference between the conditional transition probability values ​​between the two ends of each directed edge, and generate a co-occurrence probability difference matrix of adjacent nodes.

[0026] S313: Call the adjacent node co-occurrence probability difference matrix and perform a numerical comparison with the preset behavior coherence break threshold, filter out abnormal difference items in the matrix whose values ​​exceed the behavior coherence break threshold, and perform weighted cumulative summation on the abnormal difference items according to the execution sequence of the teaching activities to construct behavior transfer sparsity distribution information.

[0027] The present invention is improved in that the setting method of the behavior coherence breaking threshold is as follows: obtain a batch sample set of historical teaching activity sequences, calculate the co-occurrence probability difference between all adjacent teaching activity node pairs in the sample set, construct a statistical distribution histogram of the co-occurrence probability difference, calculate the mean and standard deviation of the difference distribution based on the histogram, and add the mean to the standard deviation by a preset multiple as the behavior coherence breaking threshold.

[0028] The process of weighted cumulative summation of abnormal difference items according to the execution sequence of teaching activities is as follows: by identifying the relative position index of abnormal difference items in the execution sequence of teaching activities, assigning decreasing weight factors according to the reverse order of time position index, calculating the position weight coefficient of each abnormal difference item, multiplying the value of each abnormal difference item by the corresponding position weight coefficient, and performing cumulative summation on the product result.

[0029] The present invention is improved in that the step of obtaining the stability break point is specifically as follows:

[0030] S411: Based on the strongly connected components in the graph structure embedded in the teaching activity, extract the continuous behavior block sequence composed of connected nodes in chronological order. For each pair of adjacent behavior blocks in the sequence, extract the high-order attribute vector representing the topological structure features. Calculate the Euclidean distance between the two high-order attribute vectors in the feature space. Connect the distance values ​​in chronological order and perform smoothing to generate a behavior density projection curve.

[0031] S412: Map the behavior density projection curve to a two-dimensional density clustering space, define the neighborhood scanning radius and minimum number of contained points for each projection point in the space, divide the projection points into core objects, boundary points or noise points according to the neighborhood density attributes of the points, calculate the density decrease rate when transitioning from a high-density core region to a low-density noise region, and establish a region density abrupt gradient.

[0032] S413: The gradient of the abrupt change in regional density is called and compared with the preset stability discrimination threshold. Abnormal fluctuation indices with gradient magnitudes exceeding the stability discrimination threshold are filtered out. The abnormal fluctuation indices are mapped back to the original time axis coordinates to lock the specific moment when the behavior pattern undergoes a qualitative change during the teaching process and obtain the stability breakpoint.

[0033] The present invention is improved in that the steps for obtaining the teaching quality level are specifically as follows:

[0034] S511: Call the target offset measurement parameters, behavior transfer sparsity distribution information and stability breakpoint, perform feature dimension standardization processing on heterogeneous data to eliminate dimensional differences, map it to the preset high-dimensional feature space coordinate system as independent component coordinates, perform vectorized concatenation on the component coordinates according to the arrangement order of feature weights, and construct teaching quality evaluation vector.

[0035] S512: Obtain a preset standard stable state vector as a reference benchmark, calculate the Euclidean distance between the teaching quality evaluation vector and the standard stable state vector in the high-dimensional feature space, and characterize the degree of deviation of the current teaching state from the ideal stable state by quantizing the magnitude of the difference vector, and generate a state vector space deviation value.

[0036] S513: The state vector space deviation value is compared with the preset multi-level quality assessment thresholds step by step to identify the specific numerical range in which the state vector space deviation value falls, and the corresponding evaluation tag is retrieved according to the preset interval index and the mapping relationship between the level definition table to obtain the teaching quality level.

[0037] A big data-based teaching quality assessment system, wherein the big data-based teaching quality assessment system is used to implement the above-mentioned big data-based teaching quality assessment method, the system comprising:

[0038] The multidimensional performance processing module obtains the original performance point value sequence of teaching tasks and the target performance interval of teaching units. It extracts continuous teaching cycle data from the original performance point value sequence of teaching tasks and constructs a multidimensional performance reference vector for the same period by combining it with the target performance interval of teaching units.

[0039] The target offset analysis module performs a difference analysis on the original performance point value sequence of the teaching task and the multidimensional performance reference vector of the same period. It uses a dynamic time warping algorithm to calculate the discrete variation amplitude within the period and calculates the ratio of the discrete variation amplitude within the period to the target performance interval of the teaching unit to generate the target offset measurement parameter.

[0040] The behavior transfer recognition module constructs a teaching activity embedding graph structure based on the original performance point value sequence of the teaching task, calculates the co-occurrence probability difference between adjacent nodes, and accumulates the co-occurrence probability difference exceeding the behavior coherence breaking threshold to generate behavior transfer sparsity distribution information.

[0041] The stability analysis module extracts continuous behavior block sequences from the embedded graph structure of the teaching activities and calculates the Euclidean distance, generates behavior density projection curves, and uses a density-based noise-based spatial clustering algorithm to extract stability breakpoints based on the behavior density projection curves.

[0042] The teaching quality classification module constructs a teaching quality evaluation vector based on the target offset metric parameter, behavior transition sparsity distribution information, and stability breakpoint, and calculates the deviation value between the teaching quality evaluation vector and the standard stable state vector to determine the teaching quality level.

[0043] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0044] In this invention, a multi-dimensional reference vector is constructed by extracting continuous teaching cycle data and combining it with the target performance interval of teaching units. The teaching performance offset parameters are generated by executing the difference and dynamic time warping algorithm. A teaching activity graph structure based on time adjacency is established, and the difference in behavior co-occurrence probability is calculated to obtain the behavior transfer sparsity distribution. A behavior density curve is generated based on the higher-order attribute Euclidean distance and the gradient of regional density mutation is identified. The teaching stability breakpoint is extracted, and multiple key features are mapped to a high-dimensional space to construct an evaluation vector and compare the deviation with the standard state. This realizes multi-dimensional dynamic modeling and stability analysis of teaching quality, effectively improving the systematicness, dynamism and objectivity of teaching quality evaluation in complex teaching scenarios. Attached Figure Description

[0045] Figure 1 This is a flowchart of the method of the present invention;

[0046] Figure 2 This is a flowchart illustrating the process of obtaining a multi-dimensional performance reference vector within the same period, as described in this invention.

[0047] Figure 3 This is a flowchart illustrating the process of obtaining target offset metric parameters according to the present invention;

[0048] Figure 4 This is a flowchart illustrating the process of obtaining behavior transfer sparsity distribution information according to the present invention.

[0049] Figure 5 This is a flowchart illustrating the process of obtaining the stability break point in this invention;

[0050] Figure 6 This is a flowchart for obtaining the teaching quality level in this invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0052] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0053] Please see Figure 1This invention provides a technical solution, a teaching quality evaluation method based on big data, comprising the following steps:

[0054] S1: Obtain the original performance point value sequence of teaching tasks and the target performance interval of teaching units through the teaching management database. Extract continuous teaching cycle data based on the timestamp attribute of the original performance point value sequence of teaching tasks. Combine the target performance interval of teaching units to construct a multi-dimensional performance reference vector for the same period.

[0055] S2: Perform a difference operation on the original performance point value sequence of the teaching task and the multi-dimensional performance reference vector of the same period, call the dynamic time warping algorithm to calculate the periodic discrete change amplitude of the original performance point value sequence of the teaching task relative to the multi-dimensional performance reference vector of the same period, and generate the target offset measurement parameter based on the ratio of the periodic discrete change amplitude to the target performance interval of the teaching unit.

[0056] S3: Based on the teaching activity nodes corresponding to the original performance point value sequence of the teaching task, construct the teaching activity embedding graph structure according to the temporal adjacency relationship of the teaching activity nodes, calculate the co-occurrence probability difference between any adjacent teaching activity nodes in the teaching activity embedding graph structure, and perform cumulative calculation on the co-occurrence probability difference exceeding the preset threshold to generate behavior transfer sparsity distribution information.

[0057] S4: Extract continuous behavior block sequences from the embedded graph structure of teaching activities, calculate the Euclidean distance between higher-order attributes in the continuous behavior block sequences and generate behavior density projection curves, input the behavior density projection curves into the density-based noise application spatial clustering algorithm to identify the region density abrupt gradient, and extract the stability breakpoints in the teaching process based on the region density abrupt gradient.

[0058] S5: Map the target offset metric parameters, behavior transition sparsity distribution information, and stability breakpoint to a high-dimensional feature space to construct a teaching quality evaluation vector, calculate the deviation between the teaching quality evaluation vector and the standard stable state vector, and determine the teaching quality level based on the deviation.

[0059] The multi-dimensional performance reference vector for the same period includes the lower bound component of the performance benchmark, the mean expected value of target achievement, and the tolerance threshold for periodic fluctuations. The target deviation measurement parameters include the polarity identifier of positive and negative deviations, the ratio of relative fluctuation amplitude, and the cumulative drift trend index. The behavior transfer sparsity distribution information includes the probability density of transfers between nodes, the cumulative weight of path breaks, and the sparse interval span distribution. The stability breakpoint includes the timestamp index of mutation occurrence, the amplitude of density gradient jump, and the label of behavior pattern conversion type. The teaching quality level includes the comprehensive stability score, the category of teaching goal achievement, and the abnormal risk warning level.

[0060] Please see Figure 2The specific steps for obtaining the multidimensional performance reference vector for the same period are as follows:

[0061] S111: Retrieve the original performance point value sequence of teaching tasks, including historical scoring records, and the preset target performance interval of teaching units from the teaching management database. Analyze the timestamp attribute carried by each discrete data item in the original performance point value sequence of teaching tasks one by one. Sort the original performance point value sequence of teaching tasks in ascending order of time dimension according to the value of the timestamp attribute. Establish the corresponding data index mapping relationship between the original performance point value sequence of teaching tasks and the target performance interval of teaching units, and generate the basic data combination of teaching tasks with timestamp attributes.

[0062] A connection channel is established with the storage partition of a specific course (e.g., "Advanced Mathematics A"). An SQL query is executed to retrieve the original performance point value sequence of the course's teaching tasks from week 1 to week 8 of the fall semester of 2024. This sequence contains a series of discrete score values ​​(e.g., 85, 88, 92, 78, etc.) and the corresponding database record creation times. Simultaneously, the system's preset target performance interval for this course unit is retrieved. In this embodiment, this interval is set as follows: This indicates the expected performance fluctuation range. Subsequently, a data parsing program is initiated to perform byte stream analysis on each discrete data item in the original performance point value sequence of the teaching task, locating and extracting date and time strings conforming to the ISO8601 standard, and converting them to Unix timestamp format (e.g., converting "2024-09-01 08:30:00" to...). The timestamp attribute is used as the data item's timestamp value. Based on this, the QuickSort algorithm is used to rearrange the entire sequence of original performance point values ​​for the teaching task in ascending order, using the timestamp value as the key, ensuring that the data items strictly follow the natural chronological distribution of the teaching activities. After sorting, the sorted sequence is traversed, and each performance point value is assigned a time-order-based index ID, along with the preset target performance range for the teaching unit. This metadata tag is attached to each data node to encapsulate the data structure and generate a basic data set of teaching tasks with timestamp attributes.

[0063] S112: Based on the combination of teaching task basic data with timestamp attributes, calculate the time interval between two adjacent timestamps in the sequence, compare the time interval with the preset teaching cycle continuity judgment threshold, if the time interval is less than or equal to the teaching cycle continuity judgment threshold, then the corresponding data node is judged as a continuous node, identify and extract the time series segment composed of continuous nodes, remove isolated time point data that do not meet the continuity requirements, and obtain a continuous teaching cycle performance data subsequence.

[0064] The specific process for setting the threshold for determining the continuity of the teaching cycle is as follows: by performing statistical distribution analysis on the time difference values ​​of all adjacent time nodes in the historical performance data sequence of the teaching unit, the mathematical expectation and standard deviation of the time difference are calculated, and the sum of the mathematical expectation and the standard deviation of the preset multiple is used as the threshold for determining the continuity of the teaching cycle.

[0065] Extract all of the sequence The timestamp value of each data point is denoted as... By performing difference operations Calculate the time interval between two adjacent timestamps to obtain the interval sequence. Then, apply the teaching cycle continuity threshold for filtering. The threshold setting process is as follows: Select the historical performance data sequence of the past 5 semesters for this teaching unit, extract a total of 500 sets of time difference samples between adjacent time nodes, and calculate the expected value of this sample set. For 24 hours, standard deviation Given a teaching cycle duration of 6 hours and a preset multiplier of 2, the threshold for determining the continuity of the teaching cycle is calculated. Hours. Retrieves the value of each time interval in the current sequence. Compare with 36 hours, if Then determine and Nodes belonging to the same coherent behavior block are marked as coherent nodes; otherwise, they are considered breakpoints. The program automatically identifies and connects all consecutive coherent nodes, extracts the longest time series segment composed of coherent nodes (e.g., the data segment from week 3 to week 5), and removes isolated time point data that do not meet the coherence requirement and have a time interval exceeding 36 hours from the sequence, thus obtaining a continuous teaching cycle performance data subsequence.

[0066] S113: Call the target performance interval of the teaching unit in the combination of the continuous teaching cycle performance data subsequence and the teaching task basic data with timestamp attribute, set the lower and upper bounds of the target performance interval of the teaching unit as the reference axis of the multi-dimensional reference coordinate system, project each performance point value in the continuous teaching cycle performance data subsequence to the multi-dimensional reference coordinate system, calculate the Euclidean space position coordinate and vertical distance parameter of each performance point value relative to the reference axis, and vectorize and concatenate the position coordinate and vertical distance parameter according to the chronological order of the time dimension to construct the multi-dimensional performance reference vector of the same period;

[0067] Construct a multi-dimensional reference coordinate system with time as the horizontal axis and performance value as the vertical axis. Define the lower bound of the interval, 75, as the lower reference axis. The upper limit of 95 is defined as the upper reference axis. For each performance point value in the sequence. (For example, take) Project it onto this coordinate system and calculate... Position coordinates in Euclidean space And calculate the perpendicular distance from that point to the lower reference axis. and the vertical distance to the upper reference axis Position coordinates With two vertical distance parameters Combined into a four-dimensional feature vector Following the chronological order of time, all within this period... The four-dimensional feature vectors at each time step are vertically concatenated to form a... The matrix structure is used to construct a multi-dimensional performance reference vector with the same period.

[0068] Please see Figure 3 The specific steps for obtaining the target offset metric parameters are as follows:

[0069] S211: Perform point-by-point difference operation based on time index on the original performance point value sequence of teaching tasks and the multi-dimensional performance reference vector of the same period, calculate the numerical deviation between the two at the same time node, and generate an instantaneous performance deviation sequence.

[0070] Align the sampling time points of both, for any time node Extract raw performance point values (For example, 88 points) and the benchmark reference value implied in the multidimensional performance reference vector of the same period. (Using the midpoint of the interval, 85 points, as the baseline), perform subtraction. Repeat this operation for all time points in the sequence to generate an instantaneous performance deviation sequence containing positive and negative deviation values.

[0071] S212: Based on the instantaneous performance deviation sequence, a minimum cumulative distance path is constructed between the original performance point value sequence of teaching tasks and the multi-dimensional performance reference vector of the same period. The overall fluctuation intensity of the sequence is calculated by integrating the path normalization cost and local deviation characteristics, using the formula:

[0072] ;

[0073] The calculation yields the discrete variation amplitude within the period;

[0074] in, This represents the magnitude of discrete changes within the period. The sequence length representing the instantaneous performance deviation sequence is obtained by counting the total number of time points contained in the sequence. The normalized instantaneous performance deviation value represents the k-th time point in the instantaneous performance deviation sequence. This value is obtained by dividing the original deviation value at the k-th time point by a preset maximum performance score and then performing a dimensionless transformation. This represents the time-decreasing weight factor for the k-th time point, calculated by applying an exponential decay function to the time index of the k-th time point. The normalized dynamic programming minimum cumulative path cost between the original performance point value sequence of the teaching task and the multi-dimensional performance reference vector of the same period is calculated by applying a dynamic time warping algorithm to the normalized sequence. The normalized teaching task performance value at the k-th time point in the original performance point sequence represents the teaching task performance value obtained by mapping the original performance value at the k-th time point to a standard unit interval. This represents the normalized intra-period reference value at the k-th time point in the multi-dimensional performance reference vector within the same period. It is obtained by mapping the reference vector value at the k-th time point to a standard unit interval. This represents the normalized adjustment coefficient for path cost;

[0075] The overall volatility intensity of the sequence is calculated by integrating path regularization costs and local bias characteristics, using the following formula:

[0076] ;

[0077] The calculation yields the discrete variation amplitude within the period;

[0078] The calculation logic and parameter settings of this formula are as follows: First, determine the sequence length. The selected continuous teaching cycle includes... Key assessment time points. Parameters For the first Normalized instantaneous performance deviation values ​​at each time point. The preset performance score is 100 points, and the original deviation values ​​for time points 1 to 5 are as follows: After normalization They are respectively .parameter To reduce the weighting factor over time, an exponential decay function is used. Calculate, assuming attenuation coefficient .but , , , , .parameter and These are the normalized teaching task performance value and the normalized reference value for the same period, respectively. The raw scores are then mapped to... The interval, from point 1 to point 5 for Corresponding (Reference benchmark) is .parameter This is to minimize the cumulative path cost using normalized dynamic programming. The minimum path distance between the original sequence and the reference sequence is calculated using the Dynamic Time Warping (DTW) algorithm. The calculated original cumulative distance is set to 1.5, and then normalized. .parameter This is the normalized adjustment coefficient for path cost, used to balance the weights of local fluctuations and overall morphological differences, and is set to 0.5.

[0079] Table 1 lists the specific values ​​and intermediate calculation results of the above parameters:

[0080]

[0081] The specific calculation is as follows: Part 1 (Locally Weighted Fluctuation Term):

[0082] ;

[0083] This section introduces a weight that decreases over time. This makes the impact of recent performance deviations on overall volatility greater than that of long-term deviations (setting a time index). A larger value represents older data, or vice versa; it's set to [value] here. (The most recent moment has the largest weight), and the root mean square operation quantifies the local dispersion after weighting.

[0084] Part Two (Overall Morphological Regularity):

[0085] ;

[0086] This section utilizes DTW cost. The overall morphological difference of the sequence under time axis distortion is measured and normalized by dividing by the total amplitude.

[0087] Final result:

[0088] ;

[0089] This numerical result It represents the discrete variation within the period after considering the time weight decay and the overall path regularization cost. The smaller the value, the more stable the performance fluctuation and the closer it is to the reference benchmark. The discrete variation within the period is calculated.

[0090] S213: Analyze the upper and lower boundary values ​​of the target performance interval of the teaching unit and calculate the interval span amplitude. Calculate the ratio of the discrete variation amplitude within the period to the interval span amplitude to obtain the target offset measurement parameter.

[0091] Analysis of the target performance range of teaching units Extract the upper boundary value of 95 and the lower boundary value of 75, and perform a subtraction operation to calculate the range amplitude. The discrete variation amplitude within the period calculated in step S212 is called. Perform a division operation to calculate the ratio of the two: This ratio maps the dimensionless fluctuation range to a specific performance evaluation interval scale, thus obtaining the target offset measurement parameter.

[0092] Please see Figure 4 The specific steps for obtaining the sparsity distribution information of behavior transition are as follows:

[0093] S311: Based on the original performance point value sequence of teaching tasks, parse the teaching activity type identifier corresponding to each performance point value, instantiate the teaching activity type identifier into an independent node in the topology network, establish directed connection edges according to the adjacent order of the teaching activity type identifier in the time dimension, and construct the teaching activity embedding graph structure using the reciprocal of the time interval between nodes as the edge weight parameter.

[0094] Tracing the metadata of teaching activities behind each performance point value, we extract the corresponding teaching activity type identifier (e.g., lecture, group discussion, experiment, quiz). These identifiers are then instantiated as independent nodes in a topological network; for example, node A represents lecture, and node B represents quiz. Directed edges are established based on the actual temporal order of the teaching activities (e.g., A before B). Calculate the time interval between node A and node B. (For example, 2 hours), take the reciprocal. The weights of the directed edges are used as parameters. All activity sequences are traversed to construct a teaching activity embedding graph structure containing nodes, directed edges, and weights.

[0095] S312: For the embedded graph structure of teaching activities, traverse all combinations of predecessor and successor nodes connected by directed edges, count the joint occurrence frequency of each node combination in the overall teaching cycle, and convert it into conditional transition probability value. Calculate the absolute value of the difference between the conditional transition probability values ​​between the two ends of each directed edge, and generate a co-occurrence probability difference matrix of adjacent nodes.

[0096] Traverse all predecessor nodes connected by directed edges (denoted as...). ) and successor node (denoted as Combine and count the number of node combinations for each group. Frequency of joint occurrences within the overall teaching cycle and predecessor nodes Total frequency of occurrence Using the formula Calculate the conditional transition probability, and similarly calculate the reverse transition probability. .

[0097] Table 2 shows the conditional transition probability data for some node pairs:

[0098]

[0099] For each directed edge, calculate the absolute value of the difference between the conditional transition probabilities between the two endpoints. For example, regarding (theoretical lectures and in-class quizzes), The differences between all node pairs are integrated to generate a co-occurrence probability difference matrix of adjacent nodes.

[0100] S313: Call the co-occurrence probability difference matrix of adjacent nodes and perform a numerical comparison with the preset behavior coherence break threshold. Filter out abnormal difference items in the matrix whose values ​​exceed the behavior coherence break threshold. Perform weighted cumulative summation on the abnormal difference items according to the execution sequence of the teaching activities to construct behavior transfer sparsity distribution information.

[0101] The specific method for setting the behavioral coherence breakdown threshold is as follows: obtain a batch sample set of historical teaching activity sequences, calculate the co-occurrence probability difference between all adjacent teaching activity node pairs in the sample set, construct a statistical distribution histogram of the co-occurrence probability difference, calculate the mean and standard deviation of the difference distribution based on the histogram, and add the mean to the standard deviation by a preset multiple as the behavioral coherence breakdown threshold.

[0102] The process of weighted cumulative summation of abnormal difference items according to the execution sequence of teaching activities is as follows: by identifying the relative position index of abnormal difference items in the execution sequence of teaching activities, assigning decreasing weight factors according to the reverse order of time position index, calculating the position weight coefficient of each abnormal difference item, multiplying the value of each abnormal difference item by the corresponding position weight coefficient, and performing cumulative summation on the product result.

[0103] The specific method for setting the threshold for behavioral coherence breakdown is as follows: Obtain a batch sample set of historical teaching activity sequences from the past 3 years, calculate the co-occurrence probability difference between all adjacent teaching activity node pairs in the sample set, obtain 1000 difference samples, construct a histogram analysis to find the mean. Standard deviation Set the preset multiplier to 2 and calculate the threshold. Items with differences greater than 0.35 in the matrix are filtered out as outliers (as shown in Table 2). (This is considered an anomaly). Identify the relative position index of the anomaly within the instructional activity execution sequence. (Set the total length of the sequence to be) The anomaly occurred in the first... (Number of positions), assigning decreasing weight factors based on the reverse order of the position indices, and calculating the position weight coefficient. (This means that the earlier anomalies occur, the higher their weight and the greater their impact on subsequent events). Multiply the anomaly difference value of 0.65 by the weight of 0.8 to get 0.52. Perform this operation on all anomalies and sum them up to construct the behavior transition sparsity distribution information.

[0104] Please see Figure 5 The specific steps for obtaining the stability break point are as follows:

[0105] S411: Based on the teaching activity embedded graph structure, traverse the strongly connected components in the graph structure, extract the continuous behavior block sequence composed of connected nodes in chronological order, extract the high-order attribute vector representing the topological structure features for each pair of adjacent behavior blocks in the sequence, calculate the Euclidean distance between the two high-order attribute vectors in the feature space, connect the distance values ​​in chronological order and perform smoothing to generate the behavior density projection curve.

[0106] The process of setting the stability discrimination threshold is as follows: select standard teaching process data marked as stable in the historical period as the benchmark sample, obtain the reference gradient value set corresponding to the benchmark sample according to the calculation logic of generating regional density change gradient, perform statistical distribution analysis on the reference gradient value set, calculate the arithmetic mean and standard deviation, and use the linear superposition result of the arithmetic mean and three times the standard deviation as the stability discrimination threshold.

[0107] The Tarjan or Kosaraju algorithm is applied to traverse the strongly connected components of the graph structure, which represent tightly cyclical behavioral patterns in teaching activities (e.g., a "lecture-practice-feedback" cycle). A sequence of consecutive behavioral blocks (Block1, Block2...) composed of connected nodes is extracted chronologically. For each pair of adjacent behavioral blocks (e.g., Block1 and Block2), a high-order attribute vector representing their topological characteristics is extracted, including node degree distribution entropy, average path length, etc., to construct a vector. and Calculate the numerical Euclidean distance between these two higher-order attribute vectors in the multidimensional feature space. The calculation yields a series of distance values. The time sequence is connected and smoothed using a moving average filter to eliminate noise interference and generate behavioral density projection curves.

[0108] S412: Map the behavior density projection curve to a two-dimensional density clustering space. Define the neighborhood scanning radius and minimum number of contained points for each projection point in the space. Divide the projection points into core objects, boundary points or noise points according to the neighborhood density attributes of the points. Calculate the density decrease rate when transitioning from a high-density core region to a low-density noise region and establish a region density abrupt gradient.

[0109] Define the neighborhood scan radius for each projection point in space. With minimum number of points The DBSCAN density clustering algorithm is executed. Based on the neighborhood density attributes of each point, projected points are divided into core objects (high-density areas), boundary points (transition areas), or noise points (low-density areas). The path from the high-density core area to the low-density noise area is identified, and the derivative of the density value with distance, i.e., the density decrease rate, is calculated. For example, if the density of a transition segment decreases from 0.9 to 0.1 at a distance of 2, the decrease rate is... This is used to establish a gradient of regional density abrupt changes.

[0110] S413: Compare the gradient of sudden change in regional density with the preset stability judgment threshold, filter out abnormal fluctuation indices whose gradient magnitude exceeds the stability judgment threshold, map the abnormal fluctuation indices back to the original time axis coordinates, lock the specific moment when the behavior pattern of teaching changes in a qualitative change, and obtain the stability breakpoint.

[0111] The process of setting the stability threshold is as follows: 100 standard teaching process data points that have been manually marked as "stable" within a historical period are selected as benchmark samples. The reference gradient value set corresponding to the benchmark samples is calculated according to the logic of S412. Statistical analysis is performed on this set, and the arithmetic mean is calculated. Standard deviation The threshold is calculated by linearly superimposing the arithmetic mean and three times the standard deviation. Filter out anomalous fluctuation indices in the current gradient with an amplitude exceeding 0.35 (e.g., points with a gradient value of 0.45), map these indices back to the original time axis coordinates, pinpoint the specific moment when the behavioral pattern undergoes a qualitative change during the teaching process (e.g., a sudden shift from intensive teaching to relaxed self-study), and obtain the stability breakpoint.

[0112] Please see Figure 6 The specific steps for obtaining the teaching quality level are as follows:

[0113] S511: Call the target offset metric parameters, behavior transfer sparsity distribution information and stability breakpoint, perform feature dimension standardization processing on heterogeneous data to eliminate dimensional differences, map it to the preset high-dimensional feature space coordinate system as independent component coordinates, perform vectorized concatenation on the component coordinates according to the arrangement order of feature weights, and construct the teaching quality evaluation vector.

[0114] By invoking the target offset metric (e.g., 0.00246), behavior transition sparsity distribution information (e.g., cumulative summation of 1.5), and stability breakpoint (e.g., the normalized position corresponding to the timestamp of 0.75), Z-score standardization is performed on the feature dimensions of these three types of heterogeneous data to eliminate dimensional differences, resulting in standardized values. Mapping these coordinates to a preset 3D feature space coordinate system as independent component coordinates, and setting the feature weights as follows: The component coordinates are weighted and vectorized according to their weight order to construct a teaching quality evaluation vector. .

[0115] S512: Obtain the preset standard stable state vector as a reference benchmark, calculate the Euclidean distance between the teaching quality evaluation vector and the standard stable state vector in the high-dimensional feature space, and characterize the degree of deviation of the current teaching state from the ideal stable state by quantifying the magnitude of the difference vector, and generate the state vector space deviation value.

[0116] Obtain the preset standard stable state vector As a reference baseline (usually set as the zero vector or the historical best state vector, set as in this example), ). Calculate the teaching quality assessment vector. The Euclidean distance between the standard stable state vector and the standard stable state vector in the high-dimensional feature space is given by the formula: By quantizing the magnitude of the difference vector, the calculation result is set. This distance value directly represents the degree of deviation of the current teaching state from the ideal stable state. The larger the distance, the higher the risk of teaching quality. This generates a deviation value in the state vector space.

[0117] S513: Compare the state vector space deviation value with the preset multi-level quality assessment thresholds step by step, identify the specific numerical range in which the state vector space deviation value falls, and retrieve the corresponding evaluation label according to the mapping relationship between the preset range index and the level definition table to obtain the teaching quality level.

[0118] The method for setting the multi-level quality assessment threshold is as follows: select a set of standard teaching tasks that have been manually rated by experts within the historical teaching cycle; calculate the historical spatial deviation value samples of each standard teaching task relative to the standard stable state vector according to the construction logic of the teaching quality assessment vector; classify and statistically analyze the historical spatial deviation value samples based on the expert rating labels; calculate the probability density function of the deviation value samples under multiple rating categories; extract the statistical quantiles at the intersection of the probability density curves of adjacent rating categories as critical limits; and sort all critical limits by numerical size to form the multi-level quality assessment threshold.

[0119] The method for setting the multi-level quality assessment thresholds is as follows: Select 1000 standard teaching tasks that have been manually rated by experts within a historical teaching cycle, categorizing them into four levels: A (Excellent), B (Good), C (Average), and D (Poor). Calculate the historical spatial deviation value samples for each task. Based on the expert rating labels, classify and statistically analyze the deviation value samples under the four rating categories, calculating the probability density function (PDF). Extract the statistical quantiles at the intersection of the probability density curves of adjacent rating categories (e.g., A and B) as the critical limits. For example, if the intersection of the PDF curves of categories A and B is 0.5, the intersection of categories B and C is 1.2, and the intersection of categories C and D is 2.0, then the threshold set is... Comparing the current deviation value of 0.8 with the threshold set reveals... If the result falls into the B category, the corresponding evaluation tag is retrieved based on the mapping relationship between the preset category index and the grade definition table, and the teaching quality grade is obtained as "Good".

[0120] A big data-based teaching quality assessment system is used to implement the aforementioned big data-based teaching quality assessment methods. The system includes:

[0121] The multidimensional performance processing module obtains the original performance point value sequence of teaching tasks and the target performance interval of teaching units. It extracts continuous teaching cycle data from the original performance point value sequence of teaching tasks and constructs a multidimensional performance reference vector for the same period by combining it with the target performance interval of teaching units.

[0122] The target offset analysis module performs a difference analysis on the original performance point value sequence of the teaching task and the multi-dimensional performance reference vector of the same period. It uses the dynamic time warping algorithm to calculate the discrete variation amplitude within the period and calculates the ratio of the discrete variation amplitude within the period to the target performance interval of the teaching unit to generate the target offset measurement parameter.

[0123] The behavior transfer recognition module constructs a teaching activity embedding graph structure based on the original performance point value sequence of the teaching task, calculates the co-occurrence probability difference between adjacent nodes, and accumulates the co-occurrence probability difference exceeding the behavior coherence breaking threshold to generate behavior transfer sparsity distribution information.

[0124] The stability analysis module extracts continuous behavior block sequences from the embedded graph structure of teaching activities and calculates Euclidean distances to generate behavior density projection curves. It then uses a density-based noise-based spatial clustering algorithm to extract stability breakpoints based on the behavior density projection curves.

[0125] The teaching quality classification module constructs a teaching quality evaluation vector based on the target offset metric parameter, behavior transition sparsity distribution information, and stability breakpoint. It then calculates the deviation between the teaching quality evaluation vector and the standard stable state vector to determine the teaching quality level.

[0126] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A teaching quality assessment method based on big data, characterized in that, Includes the following steps: S1: Obtain the original performance point value sequence of teaching tasks and the target performance interval of teaching units; extract continuous teaching cycle data from the original performance point value sequence of teaching tasks and construct a multi-dimensional performance reference vector for the same period by combining it with the target performance interval of teaching units. S2: Difference the original performance point value sequence of the teaching task with the multidimensional performance reference vector of the same period, calculate the discrete variation amplitude within the period using the dynamic time warping algorithm, and generate the target offset measurement parameter by calculating the ratio of the discrete variation amplitude within the period to the target performance interval of the teaching unit. The specific steps for obtaining the target offset metric parameters are as follows: S211: Perform a point-by-point difference operation based on time index on the original performance point value sequence of the teaching task and the multidimensional performance reference vector of the same period, calculate the numerical deviation between the two at the same time node, and generate an instantaneous performance deviation sequence. S212: Based on the instantaneous performance deviation sequence, construct the minimum cumulative distance path between the original performance point value sequence of the teaching task and the multidimensional performance reference vector of the same period, calculate the overall fluctuation intensity of the sequence by integrating the path regularization cost and local deviation characteristics, and calculate the discrete variation amplitude within the period. S213: Analyze the upper and lower boundary values ​​of the target performance interval of the teaching unit and calculate the interval span amplitude, calculate the ratio of the discrete variation amplitude within the period to the interval span amplitude, and obtain the target offset measurement parameter; S3: Construct a teaching activity embedding graph structure based on the original performance point value sequence of teaching tasks, calculate the co-occurrence probability difference between adjacent nodes, and accumulate the co-occurrence probability difference exceeding the behavioral coherence breaking threshold to generate behavioral transfer sparsity distribution information. The specific steps for obtaining the behavior transfer sparsity distribution information are as follows: S311: Based on the original performance point value sequence of teaching tasks, parse the teaching activity type identifier corresponding to each performance point value, instantiate the teaching activity type identifier into an independent node in the topology network, establish directed connection edges according to the adjacent order of the teaching activity type identifier in the time dimension, and construct the teaching activity embedding graph structure using the reciprocal of the time interval between nodes as the edge weight parameter. S312: For the embedded graph structure of the teaching activity, traverse all combinations of predecessor and successor nodes connected by directed edges, count the joint occurrence frequency of each node combination in the overall teaching cycle, convert it into conditional transition probability value, calculate the absolute value of the difference between the conditional transition probability values ​​between the two ends of each directed edge, and generate a co-occurrence probability difference matrix of adjacent nodes. S313: Call the adjacent node co-occurrence probability difference matrix and perform a numerical comparison with the preset behavior coherence break threshold, filter out abnormal difference items in the matrix whose values ​​exceed the behavior coherence break threshold, and perform weighted cumulative summation on the abnormal difference items according to the execution sequence of the teaching activities to construct behavior transfer sparsity distribution information. S4: Extract continuous behavior block sequences from the embedded graph structure of the teaching activities and calculate the Euclidean distance to generate behavior density projection curves. Use density-based noise to apply spatial clustering algorithms and extract stability breakpoints based on the behavior density projection curves. The specific steps for obtaining the stability break point are as follows: S411: Based on the strongly connected components in the graph structure embedded in the teaching activity, extract the continuous behavior block sequence composed of connected nodes in chronological order. For each pair of adjacent behavior blocks in the sequence, extract the high-order attribute vector representing the topological structure features. Calculate the Euclidean distance between the two high-order attribute vectors in the feature space. Connect the distance values ​​in chronological order and perform smoothing to generate a behavior density projection curve. S412: Map the behavior density projection curve to a two-dimensional density clustering space, define the neighborhood scanning radius and minimum number of contained points for each projection point in the space, divide the projection points into core objects, boundary points or noise points according to the neighborhood density attributes of the points, calculate the density decrease rate when transitioning from a high-density core region to a low-density noise region, and establish a region density abrupt gradient. S413: The region density mutation gradient is compared with the preset stability discrimination threshold. Abnormal fluctuation indices with gradient magnitudes exceeding the stability discrimination threshold are filtered out. The abnormal fluctuation indices are mapped back to the original time axis coordinates to lock the specific moment when the behavior pattern undergoes a qualitative change during the teaching process and obtain the stability breakpoint. S5: Construct a teaching quality evaluation vector based on the target offset metric parameters, behavior transition sparsity distribution information, and stability breakpoint, and calculate the deviation between the teaching quality evaluation vector and the standard stable state vector to determine the teaching quality level.

2. The teaching quality evaluation method based on big data according to claim 1, characterized in that, The same-period multidimensional performance reference vector includes the lower bound component of the performance benchmark, the mean of the expected target achievement, and the tolerance threshold for periodic fluctuations. The target offset measurement parameters include positive and negative deviation polarity indicators, relative fluctuation amplitude ratio, and cumulative drift trend index. The behavior transfer sparsity distribution information includes the inter-node transfer probability density, the cumulative weight of path breakage, and the sparse interval span distribution. The stability breakpoint includes the timestamp index of the mutation occurrence, the amplitude of the density gradient jump, and the behavior pattern conversion type label. The teaching quality level includes the comprehensive stability score, the teaching goal achievement category, and the abnormal risk warning level.

3. The teaching quality evaluation method based on big data according to claim 2, characterized in that, The specific steps for obtaining the multi-dimensional performance reference vector for the same period are as follows: S111: Retrieve the original performance point value sequence of teaching tasks, including historical scoring records, and the preset target performance interval of teaching units from the teaching management database. Analyze the timestamp attribute carried by each discrete data item in the original performance point value sequence of teaching tasks one by one. Sort the original performance point value sequence of teaching tasks in ascending order of time dimension according to the value of the timestamp attribute. Establish the corresponding data index mapping relationship between the original performance point value sequence of teaching tasks and the target performance interval of teaching units, and generate a combination of basic data for teaching tasks with timestamp attributes. S112: Based on the combination of teaching task basic data with timestamp attributes, calculate the time interval value between two adjacent timestamps in the sequence, compare the time interval value with the preset teaching cycle continuity judgment threshold, if the time interval value is less than or equal to the teaching cycle continuity judgment threshold, then the corresponding data node is judged as a continuous node, identify and extract the time sequence segment composed of continuous nodes, remove isolated time point data that do not meet the continuity requirements, and obtain a continuous teaching cycle performance data subsequence. S113: Call the target performance interval of the teaching unit in the combination of the continuous teaching cycle performance data subsequence and the teaching task basic data with timestamp attribute, set the lower and upper bounds of the target performance interval of the teaching unit as the reference axis of the multi-dimensional reference coordinate system, project each performance point value in the continuous teaching cycle performance data subsequence to the multi-dimensional reference coordinate system, calculate the Euclidean space position coordinate and vertical distance parameter of each performance point value relative to the reference axis, and vectorize and concatenate the position coordinate and vertical distance parameter according to the chronological order of the time dimension to construct the multi-dimensional performance reference vector of the same period.

4. The teaching quality evaluation method based on big data according to claim 1, characterized in that, The specific formula for obtaining the discrete variation amplitude within the period is as follows: ； in, This represents the magnitude of discrete changes within the period. The sequence length representing the instantaneous performance deviation sequence. This represents the normalized instantaneous performance deviation value at the k-th time point in the instantaneous performance deviation sequence. This represents the weighting factor that decreases over time at the k-th time point. The normalized dynamic programming approach minimizes the cumulative path cost between the original performance point sequence of a teaching task and the multidimensional performance reference vector within the same period. The normalized performance value of the teaching task at the k-th time point in the original performance point sequence of the teaching task represents the teaching task performance value. This represents the normalized intra-period reference value at the k-th time point in the intra-period multi-dimensional performance reference vector. This represents the normalized adjustment coefficient for path cost.

5. The teaching quality evaluation method based on big data according to claim 1, characterized in that, The specific method for setting the behavioral coherence break threshold is as follows: obtain a batch sample set of historical teaching activity sequences, calculate the co-occurrence probability difference between all adjacent teaching activity node pairs in the sample set, construct a statistical distribution histogram of the co-occurrence probability difference, calculate the mean and standard deviation of the difference distribution based on the histogram, and add the mean to the standard deviation by a preset multiple as the behavioral coherence break threshold. The process of weighted cumulative summation of abnormal difference items according to the execution sequence of teaching activities is as follows: by identifying the relative position index of abnormal difference items in the execution sequence of teaching activities, assigning decreasing weight factors according to the reverse order of time position index, calculating the position weight coefficient of each abnormal difference item, multiplying the value of each abnormal difference item by the corresponding position weight coefficient, and performing cumulative summation on the product result.

6. The teaching quality evaluation method based on big data according to claim 1, characterized in that, The specific steps for obtaining the teaching quality level are as follows: S511: Call the target offset measurement parameters, behavior transfer sparsity distribution information and stability breakpoint, perform feature dimension standardization processing on heterogeneous data to eliminate dimensional differences, map it to the preset high-dimensional feature space coordinate system as independent component coordinates, perform vectorized concatenation on the component coordinates according to the arrangement order of feature weights, and construct teaching quality evaluation vector. S512: Obtain a preset standard stable state vector as a reference benchmark, calculate the Euclidean distance between the teaching quality evaluation vector and the standard stable state vector in the high-dimensional feature space, and characterize the degree of deviation of the current teaching state from the ideal stable state by quantizing the magnitude of the difference vector, and generate a state vector space deviation value. S513: The state vector space deviation value is compared with the preset multi-level quality assessment thresholds step by step to identify the specific numerical range in which the state vector space deviation value falls, and the corresponding evaluation tag is retrieved according to the preset interval index and the mapping relationship between the level definition table to obtain the teaching quality level.

7. A teaching quality evaluation system based on big data, characterized in that: The system is used to implement the big data-based teaching quality assessment method according to any one of claims 1-6, and the system comprises: The multidimensional performance processing module obtains the original performance point value sequence of teaching tasks and the target performance interval of teaching units. It extracts continuous teaching cycle data from the original performance point value sequence of teaching tasks and constructs a multidimensional performance reference vector for the same period by combining it with the target performance interval of teaching units. The target offset analysis module performs a difference analysis on the original performance point value sequence of the teaching task and the multidimensional performance reference vector of the same period. It uses a dynamic time warping algorithm to calculate the discrete variation amplitude within the period and calculates the ratio of the discrete variation amplitude within the period to the target performance interval of the teaching unit to generate the target offset measurement parameter. The behavior transfer recognition module constructs a teaching activity embedding graph structure based on the original performance point value sequence of the teaching task, calculates the co-occurrence probability difference between adjacent nodes, and accumulates the co-occurrence probability difference exceeding the behavior coherence breaking threshold to generate behavior transfer sparsity distribution information. The stability analysis module extracts continuous behavior block sequences from the embedded graph structure of the teaching activities and calculates the Euclidean distance, generates behavior density projection curves, and uses a density-based noise-based spatial clustering algorithm to extract stability breakpoints based on the behavior density projection curves. The teaching quality classification module constructs a teaching quality evaluation vector based on the target offset metric parameter, behavior transition sparsity distribution information, and stability breakpoint, and calculates the deviation value between the teaching quality evaluation vector and the standard stable state vector to determine the teaching quality level.

Images